Small-Scale Study of Siberian Biomass Burning : I . Smoke Microstructure

Reliable assessment of the impact of Siberian boreal forest wildfires on the environment and climate necessitates an improved understanding of microphysical and chemical properties of emitted aerosols. Smoldering, flaming and mixed fires of typical Siberian biomass (pine and debris) were simulated during a small-scale study in a Large Aerosol Chamber (LAC). Individual particle analysis of PM10 and PM2.5 smoke morphology and elemental composition revealed a strong dependence on combustion temperature, i.e., a dominant abundance of soot agglomerates versus roughly spherical organic particles in the flaming and smoldering phase, respectively. Cluster analysis of smoke microstructure was used to apportion the emitted particles into major characteristic groups: Soot and Organic, which accounted for around 90% and 60% of total particle numbers emitted from the flaming and smoldering fires, respectively. Carbon fractions and inorganic ion analysis supported the identification of particle types representative of combustion phase and biomass type. Elemental carbon (EC) particles from flaming fires comprised approximately 25% of Group Soot, in good agreement with a high EC fraction in total carbon of around 65% and low organic carbon (OC)/EC ratio near 0.5. Smoldering fires of pine and debris produced exclusively organic particles with high OC/EC ratios of 194 and 34, respectively. Small quantities of elemental constituents in biomass were vaporized during combustion and produced internally/externally mixed fly ash in Group Ca-, Si-, and Fe-rich of significantly less abundance. Ca, Cl, S, and Mg were more frequently distributed elements in pine than debris smoke. Sulfates and nitrates produced from gas-to-particle reactions formed Group Sand N-rich. During time evolution of smoke volatile inorganic compounds were condensed as potassium chlorides and sulfates into a newly formed Group K,Cl-rich. Quantification of Siberian biomass smoke microstructure by chemical micromarkers enables aerosols to be classified with respect to a source type assigned to Siberian wildfires.


INTRODUCTION
Biomass burning and fossil fuel combustion produce large amounts of smoke in the atmosphere on global scale, affecting air quality, the earth's radiation budget, visibility, and aerosol/cloud/climate interactions (Ito and Penner, 2005).Black carbon (BC) in combustion aerosols is of critical importance among other climate-active species with respect to its large magnitude of induced temperature changes.
Hygroscopic properties of particulate matter (PM) derived from combustion processes are related to indirect impacts of smoke emissions on haze and cloud microphysics (Popovicheva, 2010;Yun et al., 2013).At the local level, combustion emissions are considered as a dangerous pollutant, leading to exacerbate respiratory and allergic diseases (Bernstein et al., 2004).Especially, biomass burning (BB) may profoundly affect the air quality and public health in urban areas, rendering smoke aerosol a major component of harmful pollution (Amiridis et al., 2012;Tao et al., 2013;Popovicheva et al., 2014a).
Aerosol generated from wildfires is the predominant type of aerosol in the Arctic atmosphere during the summer months.Quantification of Siberian atmospheric pollution and impacts on the aerosol/climate system of Arctic regions is presently one of the most important research priorities (Quinn et al., 2008).Siberian wildfires are a major source of climate-relevant species emitted at northern latitudes (Lavoué et al., 2000).Significant influence of boreal forest fire smoke on variability of BC in the atmosphere was shown by Kozlov et al. (2008).Special concern is directed on biomass and diesel combustion sources which degrade the Arctic air quality by depositing BC on snow and melting icecaps (Matsui et al., 2011).Seasonal BB activities in Siberia may double the climate-relevant species, including BC and organic aerosols (Warneke et al., 2010), and cause the pronounced pollution event of "Arctic smoke" (Stock et al., 2012).Despite the concerns about environmental significance of millions of tons of PM emitted by wildfires, systematic observations in Russia were performed only in a few places in Western and North-Eastern Siberia (Kozlov et al., 2008;Paris et al., 2009).Thus, the assessments of Siberian BB emissions and their attribution to potential adverse effects are very limited.
Radiative effects, transport, deposition, and cloud-forming potential of wildfire emissions are closely related to aerosol physico-chemical properties.The combustion phase (open flaming, smoldering or both phases mixed simultaneously in an adjacent area), the type of fire (crown or ground), biomass fuel, its moisture content, and nature of soil determine the smoke particle composition (Reid et al., 2005;Cahill et al., 2008;Chen et al., 2010).In recent investigations of prescribed burns in pine and mixed wood Siberian forests the fractions of OC and EC, ranging from 70 to 90% and 2 to 15% of PM mass, respectively, have been shown to be dominant aerosol components, while trace elements accounted for up to 8% of PM mass (Samsonov et al., 2012).The largest emissions of organics and potassium were observed in Sub-Arctic boreal forest prescribed burns in the smoldering and flaming phase, respectively (Cahill et al., 2008).However, given that large regional smoke emissions are a result of numerous fires, and the relative amount of flaming versus smoldering burns is unclear, the impacts of fresh smoke and its evolution during aging processes remain rather uncertain.
Biomass is a complex heterogeneous mixture of organic (cellulose, hemicelluloses, and lignin) and inorganic matter, containing various solid and fluid intimately associated phases.Organic minerals such as calcium oxalates, mineral species from different classes (silicates, oxyhydroxides, sulphates, phosphates, carbonates, chlorides, nitrates) of both authigenic (biomass-accumulated) and detrital origin comprise the inorganic constituents of biomass (Vassiliev, et al., 2012).According to fundamentals of biomass burning, smoke is produced from thermal decomposition of biomass in condensation processes of volatilized organic and inorganic compounds (Reid et al., 2005).During high-temperature flaming combustion, carbonaceous fuel degrades into small radicals from which aromatic rings are generated (Wang, 2011).They produce the disordered graphitic microstructure of elemental carbon in the soot particles (Postfai et al., 2003), similar to those originating from fossil fuel combustion (Demirdjian et al., 2007).Far from the flame front the primary soot particles coagulate into chain agglomerates, volatile organic, and vaporized inorganic species condense on the particle surface or form internally mixed particles of fly ash.
Smoldering is a surface process that begins when most of the volatiles have been expelled from the fuel at temperatures lower than those during the flaming phase and oxygen diffuses to the surface and reacts exothermally.Lowvolatility organic compounds account for a significant part of condensation nuclei because of their ability to undergo gas-to-particle conversion and formation of organic aerosol particles without distinct morphology (Hand et al., 2005).
For characterization of multicomponent aerosols at microscopic level individual particle analysis methods have been developed (Osán et al., 2002;Posfai et al., 2003;Hand et al., 2005).Based on composition and morphology, three distinct types of carbonaceous particles (soot, organic with K salt inclusions, and tar balls) were identified in smoke plumes from savanna BB (Posfai et al., 2003).Cluster analysis is presently a well-developed multivariate individual particle analysis technique which is used for the particle assignments into several groups with similar composition (Van Borm and Adams, 1988;Xie et al., 2005).Cluster analysis has improved BB smoke classification specifying elevated numbers of K-S and K-Cl particles, and of mineral, Si-rich, and low-Z element particles from flaming and smoldering savanna fires, respectively (Liu et al., 2000).Individual particle analysis during wildfire event revealed the multicomponent structure of smoke consisting of carbonaceous particles (soot/tar and fly ash) externally mixed with dust (Popovicheva et al., 2014a).The abundance of the group containing large soot agglomerates surrounded by organic material and tar balls has approached at least half of total aerosol concentration during an extreme wildfire event.The high abundance of this group is well correlated with elevated OC and EC mass concentrations on days with high smoke levels; therefore the Group Soot/tar was selected as a micromarker of wildfires in smoke microstructure.It was observed with high abundance in aged biomass burning aerosols after long-range transport from large scale wildfires (Diapouli et al., 2014).
Various small-scale fires in combustion chambers have been conducted for characterization of BB emissions under controlled conditions (Chen et al., 2007;Iinuma et al., 2007;Hopkins et al., 2007).This approach allows to limit the number of uncertainties in studies of wildfires, resulting from the mixture of fuels, impact of soil, wind, and unpredictable combustion phase.The morphology and composition of combustion particles from eight different wildland fuels were analyzed by Chakrabarty et al. (2006).Classification of individual particles was performed for twelve biomass species from mid-latitude forests by Hopkins et al. (2007).However, systematic investigations of Siberian boreal forest fire emissions are still missing, especially in terms of the quantification of smoke microstructure and its evolution over time.
This paper is devoted to the study of smoke microstructure of small-scale fires of typical Siberian pine wood and forest debris biomass.Smoldering, flaming, and mixed burns were conducted in a Large Aerosol chamber (LAC), followed by examination of the resulting smoke particles by scanning electron microscopy (SEM) coupled with energy-dispersion X-ray (EDX) spectroscopy, which was used to analyze the morphology and composition of individual particles.Quantification of smoke chemical microstructure was supported by cluster analysis in order to separate aerosols into characteristic groups representative of the combustion phase and biomass type.The smoke samples were also analyzed for carbon fractions and ion content to support the identification of major types of particles in each emission type.Time evolution of smoke microstructure was investigated to reveal the dependence of the aging process on combustion conditions, thus providing fundamental knowledge about the physico-chemical mechanism of aerosol formation and transformation in biomass burning smoke.

Experimental Fire Design, On-line Measurement and Sampling
To simulate the combustion phases under controlled conditions the small-scale fires were conducted in a Large Aerosol Chamber (LAC) with a total volume of 1800 m 3 , which previously was used to study the optical properties of wood smoke (Kozlov et al., 1993(Kozlov et al., , 1996)).Two ovens were located in the center of the LAC.In one oven low temperature combustion was conducted by igniting the fuel at a temperature approaching 400°C.The smoldering phase was maintained by limiting excess oxygen while visually ensuring an absence of open flames.In the other oven, open flaming burns were ignited at temperatures near 700°C.
The burning experiments were designed to simulate the individual combustion event that proceeded from ignition through either flaming or smoldering phase.This approach is advanced in comparison with those commonly used in chamber studies (Chakrabarty et al., 2006, Hopkins et al., 2007) where simulations of natural fires were performed with initial flaming followed by smoldering phase, resulting in smoke particles being collected from a mixture of combustion phases.In a few LAC experiments mixed phase combustion was also simulated by simultaneous use of both ovens while burning half of the total mass of a given fuel under smoldering and the other half under flaming conditions.
The fuels used were biomass species from closed-canopy coniferous forest, which is most typical boreal wildland forest in Siberia.Scots pine wood and forest debris (mixture of pine needles, branches, and cones) were dried during indoor storage until a fuel moisture content 12.4% and 6.8% of dry mass was reached for debris and pine wood, respectively, representing typical Siberian wildland fuels during drought episodes.Fuel moisture was determined by heating a fuel sample to 100°C and measuring the mass loss.About 400 g of pine wood and 200 g of debris was typically used for each controlled burn.The smoke from each burn was allowed to fill the chamber and remain there for a period of 48 h, in order to study the smoke time evolution.The chamber was closed during this smoke evolution time.Several (3-6) replicate burns for each fuel were performed to investigate the burn-to-burn variability of the combustion process.Outside air was pumped through the chamber prior to each burn to remove remaining smoke from previous burns.No any light was used during LAC studies to neglect by any photo-chemical processes during fire and smoke evolution.
Two sampling ports were positioned near the ovens and passed through the wall of the chamber to the laboratory where the sampling and monitoring equipment was located.PM 10 and PM 2.5 size-selective inlets (Digital) were used to sample smoke particles with aerodynamic diameters of less than 10 and 2.5 µm, respectively, at a controlled flow rate of 1 m 3 /h.Impactors with an aerodynamic cutoff diameter of 1.08 µm were utilized to collect particles on metal substrates (Cu foil) behind each inlet.Additionally, two samplers were used for sampling PM 10 and PM 2.5 mass on 47 mm quartz fiber filters (Pallflex) for measurements of OC, EC, and water-soluble ions.Before sampling, the quartz fiber filters were pre-treated at 500°C for 6 hours.Sets of PM 10 and PM 2.5 samples were collected from replicate burns for each biomass type (pine wood and debris) and combustion phase (smoldering, flaming, and mixed).PM mass of all samples was determined gravimetrically by pre and post weighting the filters under controlled humidity in desiccators.
A polar spectronephelometer (APSN-02) was used for on-line aerosol scattering measurements, as described in (Rakhimov et al., 2014).After ignition the aerosol scattering, which was related to smoke particle mass concentration, was continuously rising and approached the maximum levels typically in 2 hours, when smoke was dispersed and had filled the chamber homogeneously.At this time the sampling of fire-emitted particles was commenced; the smoke particles collected at this point were term "fresh".To evaluate the time evolution, smoke was able to stay in LAC during one-two days more, it was called as "aged".Sampling was performed 24 h and 48 h after fire ignition when the smoke mass concentration in the chamber had significantly decreased.All samples were stored in a refrigerator at 4°C prior to chemical analyses in the laboratory.

Individual Particle Analysis
Samples for each combustion phase and biomass species were examined using a LEO 1430-vp (Karl Zeiss) field emission scanning electron microscope with a spatial resolution of 7 nm, equipped with an Oxford energy dispersive detector INCA.Energy dispersive X-ray spectroscopy spectra for Z elements (Z ≥ 5) were recorded in SEM image mode and then quantified by means of a method that relates the measured X-ray intensity to the elemental concentration, using calculated equivalent X-ray intensities of corresponding elements.Samples were studied in the high vacuum mode at 10 kV acceleration voltage and a beam current of 1 nA.Counting time for X-ray spectra was 50 seconds.
The typical SEM analysis started with the collection of EDX spectra from several areas of a blank Cu foil substrate for control of its homogeneity.The small concentrations of C and O were subtracted from the EDX spectra of sampled particles.The analysis area of each sample was subdivided into different fields if particle morphology differences were clearly observed.Approximately 500 individual particles with a diameter from 0.1 up to 2.5 µm and 10 µm were measured in each PM 2.5 and PM 10 sample of repeated burns, respectively.This number was considered to be sufficient for obtaining a representative overview of the individual elements and groups in each sample (Liu et al., 2000;Chakrabarty et al., 2006;Popovicheva et al., 2012Popovicheva et al., , 2014a)).Altogether about 4700 particles and 14 elements were analyzed in the collected samples.Particles smaller than 0.1 µm were ignored because their image cannot be obtained with sufficient resolution when the X-ray spectra are accumulated.Since highly irregular and agglomerated shapes of particles prevent the size distribution to be obtained from SEM images, only averaged sizes of primary particles in chain agglomerates were estimated using their projected area equivalent diameters.
EDX analysis yields a data matrix containing the element weight concentrations for C, O, Сa, Al, Si, S, Cl, K, Fe, Ti, P, Na, Mg, and N in the smoke particles.A combination of hierarchical and non-hierarchical cluster analysis, k-means and g-means, was applied for separation of individual particles into characteristic groups of similar chemical composition using the software Deductor.Details of the theoretical approach are described elsewhere (Popovicheva et al., 2012).Success of clustering relies on the selection of the correct number of groups representing the data.After completion of the clustering step, groups with an average composition as close as possible to physico-chemically identifiable particle types are chosen.In this case, the clustering with a lower number of groups does not provide good separation, while a higher number leads to duplication of groups with similar composition.The naming of particle groups is based on both morphological features and the most abundant elements (after C and O).

Bulk Chemical Analysis
Bulk chemical analysis of carbon fractions and ion content was performed to support the identification of particle types in the different groups (Niemi et al., 2006;Bladt et al., 2012;Popovicheva et al., 2012).OC and EC were determined by the NIOSH protocol with a maximum temperature of 840°C in the He-mode (NIOSH, 2003), by thermal-optical transmittance (TOT), using a Sunset carbon analyzer (Sunset Laboratory, Inc.).Total carbon (TC) was obtained as the sum of OC and EC.Carbonate Carbon (CC) was determined by manual integration of the sharp peak (whenever present) occurring during the transition to the maximum temperature step in the inert mode (Karanasiou et al., 2011).
The inorganic anions and cations, including sulfate ), sodium (Na + ), potassium (K + ), calcium (Ca 2+ ), and ammonium (NH 4 + ), were measured by ion chromatography (IC) with conductivity detection, using a Dionex ICS-3000 IC system (Thermo Scientific, Sunnyvale, CA, USA) (Zhang et al., 2013).Filter samples were cut into 17 mm round portions, extracted in deionized ultra-pure water under ultrasonic agitation for one hour, and filtered (using PTFE syringe filters).The cation separation was achieved on a Dionex IonPac CS12 column with a 20 mM methanesulfonic acid eluent at a flow rate of 1.0 mL/m.Anions were separated on a AS14 column with an eluent composed of 1.7 mM NaHCO 3 and 1.8 mM Na 2 CO 3 at a flow rate of 1.2 mL/m.

Morphology and Elemental Composition
The SEM panorama of pine wood smoke sampled in the flaming phase shows an impaction spot of concentricallyoriented particles (Fig. 1(a)).The particle images present two morphological types: chain and crystalline.The major type of particles is soot, which can be clearly distinguished by the agglomerates of ultrafine primary particles containing from a few up to hundreds of spheres (Fig. 1(c)).The average size of primary particles was found to be around 90 nm.Fly ash particles in the flaming phase demonstrate the solid irregular shapes of round, elongated, and euhedral morphology.
Dense particle-phase compounds from smoldering burns produce a quasi-liquid substance at the center of a spot with a wide layer of spilled particles at its boundary (Figs.1(b) and 1(d)).EDX analysis of the spot center shows 80 wt% of C and 20 wt% of O, the composition known for pine tar.Particles accumulated at the spot boundary demonstrate either liquid-like roughly spherical or euhedral morphology.With decreasing smoke concentration in the chamber pine tar was no longer observed on the impaction spot.
The major elements in various biomass types, in decreasing order of abundance, are commonly C, O, H, N, Ca, K, Si, Mg, Al, S, Fe, P, Cl, Na, Mn and Ti (Vassiliev, et al., 2012).Practically all are found in fresh smoke from various biomass types (Reid et al., 2005).The composition of individual Siberian biomass smoke particles was found to vary over a wide range of C and O, as well as 14 trace elements.Nearly every particle in pine and debris smoke was found to be carbonaceous, i.e., its major elements were C and O.The abundance of trace elements in smoke particles is presented in Fig. 2, indicating Ca, Si, Al, S, K, Mg, and Cl as the most frequently distributed elements.In flaming and smoldering fires from 3 to 12% and from 5 to 22% of smoke particles, respectively, contain these elements.Ca, Cl, Mg, Si, Al, and S were found to be more abundant in pine than debris smoldering smoke particles.Interestingly, time evolution of flaming smoke in the chamber lead to a prominent increase in the abundance of K, Cl, and S in particles, by factors of three, two, and five, respectively, while they remained nearly constant for smoldering smoke evolution in the chamber, except for S.

Clustering and Smoke Microstructure
Individual particle analysis revealed the complex morphological and chemical composition of smoke particles produced by small-scale fires of Siberian biomass.At microscopic level chemical components were found to be distributed heterogeneously throughout the particles.SEM/EDX measurement showed carbonaceous particles to be internally and externally mixed with inorganic fly ash.Groups of particles from smoldering and flaming fires obtained by cluster analyses are presented in Table 1 for fresh and aged smoke (with 48 h evolution) in the chamber.We found a small difference between the abundance of  groups in PM 10 and PM 2.5 samples, just a few percents for debris smoke in the flaming phase, as shown in Table 1.The concentration of coarse particles in smoke was low, indicating that their composition cannot influence the group separation significantly.PM 10 and PM 2.5 mass, averaged over replicate burns for each biomass type and combustion phase as well as carbon fractions in PM are shown in Table 2. Since no significant difference between carbon fractions in PM 2.5 and PM 10 was found, all data are reported from hereon for PM 2.5.The PM ion content is shown in Table 3 and gives a good overview of the bulk composition of water-soluble inorganic compounds in smoke particles.
In smoke from flaming fires the group of chain agglomerates of roughly spherical primary particles contained mainly C and O with negligible amounts of other elements (< 3 wt%); these particles are assigned to Group Soot.In pine and debris smoke, the Group Soot comprised 87 and 92% of total particle numbers, which is in good agreement with the TC content of 59 and 47% in PM mass, respectively.Individual particle analysis firstly demonstrated the particles of pure elemental carbon composition.Around 25% of all particles in Group Soot contained 100% C; their typical SEM micrograph is shown in Fig. 3.1.The remaining 75% particles in Group Soot were found to contain 3-6% of O (Fig. 3.2).Such oxygen content is typical for soot produced by hydrocarbon and fossil fuel combustion (Kireeva et al., 2009;Bladt et al., 2012), and determinesthe hydrophobic character of soot interaction with water (Popovicheva et al., 2008).Bulk analysis confirmed the high EC fraction (68 and 63% in TC) with OC/EC ratios of 0.46 and 0.57 for pine and debris smoke, respectively.

Group
In previous studies perfectly spherical carbonaceous particles were observed in the smoldering phase of fires and have been termed ''tar balls'', which are believed to be formed by bimolecular homogeneous nucleation of organic matter with water vapor (Posfai et al., 2004;Hand et al., 2005;Chakrabarty et al., 2006).However, the distinction between organic particles and tar balls is highly uncertain.In the LAC camber burns, we observed only aluminosilicates as perfectly spherical particles, possibly because fuels were too dry to produce tar balls.
Mixed combustion yielded both soot agglomerates and organic particles in C and O-containing group, as it was observed in simulations of natural fires with initial flaming following by smoldering (Chakrabarty et al., 2006, Hopkins et al., 2007).The ratio of OC/EC for mixed fires was obtained ~6.7 in our study, well between those obtained in the smoldering and flaming phases (Table 2).Similar OC/EC ratios near 10 were obtained during prescribed burns in a Siberian coniferous forest (Samsonov et al., 2012).
In fly ash Groups Ca, Si, Fe, and S-rich contained Ca, Si, Fe and Si elements as most abundant after C, while the other trace elements were found with variable composition range, (Table 1).In the flaming phase, the Group Ca-rich in pine and debris smoke comprised 5.6 and 1.4% of total particle number, while in the smoldering phase it approached 15.0 and 11.4%, respectively.Fig. 4.1 shows the typical SEM micrograph and EDX spectra of Group Ca-rich particles dominated by Ca with minor inclusions of S, Cl, Mg, P, and Si.The biggest part of Group Ca-rich particles contained only Ca (Fig. 4.2), probably in form of lime (CaO) or calcium hydroxide (Ca(OH) 2 ).In addition, particles of this group may have been produced by evaporation of Ca oxalates from the biomass structure as CaC 2 O 4 •H 2 O was identified in low-temperature ash of different biomass species (Suarez-Garcia et al., 2002).Carbonate (CaCO 3 ) was proposed as the second major type of particles in Group Ca-rich, because carbonate carbon was measured in some PM 10 samples at concentrations up to 0.7%.The detrital presence of calcite in biomass from soil has also been mentioned (Vassilev et al., 2012).The mineral composition of the biomass ashes clearly showed the intensive formation of various newly formed carbonates (calcite and dolomite) as a result of gasto-particle reactions between the volatile CO 2 released from biomass and alkaline-earth and alkaline oxyhydroxides (Suarez-Garcia et al., 2002).Since Ca 2+ ions were measured in the flaming phase and were absent in the smoldering phase (Table 3), we may assume the formation of calcium carbonates occurred during high-temperature combustion of Siberian biomass.
Approximately with the same abundance as Group Carich an additional Group Si-rich was observed in pine and debris smoke with similar composition from both flaming and smoldering fires.Silicates of both detrital and authigenic origin are typical component of biomass (Vassiliev et al., 2012).Opal (SiO 2 •nH 2 O) plays a role for structural stability Fig. 4. Representative SEM micrographs and EDX spectra of particles in Group Ca-rich 1) CaSO 4 /CaCl 2 , 2) CaO/CaCO 3 ; in Group Si-rich, 3) Ca-aluminosilicates with S and Cl, 4) quartz; in Group Fe-rich, 5) Fe-aluminosilicates with K, Mg; in Group S-rich, 6) H 2 SO 4 /(NH 4 ) 2 SO 4 , 7) K 2 SO 4 ; in Group K, Cl-rich, 8) KCl (Cu is a substrate artifact).of plants, typical soil minerals (feldspars clay, mica, quartz) and are introduced by wind and deposited on plant surfaces.Therefore, we may relate the observed composition of Group Si-rich to soil mineral microcrystallites among the inorganic constituents of pine and debris biomass.Fig. 4.3 shows Ca-dominated aluminosilicates, partially mixed with S, K, and Cl from debris smoldering fires.The other part of Group Si-rich particles was assigned to quartz because of the nearly exclusive Si and O content, as shown in Fig. 4.4 for the pine flaming phase.Group Fe-rich was found with the smallest abundance only in the smoldering phase, and was composed of Fe-dominated aluminosilicates mixed with K and Mg (Fig. 4.5).
Sulphates, chlorides, and phosphates have been identified in biomass as mobile mineral class of authigenic origin due to precipitation of water in natural biomass (Vassiliev et al., 2012).Sulphate formation during combustion is attributed to gas-to-particle reactions of acidic SO 2 and SO 3 gases, formed from biomass S-containing compounds, and basic alkaline and alkaline-earth ions bound to organic matter of biomass as exchangeable elements, associated with arcanite (K 2 SO 4 ) and anhydrite (CaSO 4 ) (Dare et al., 2001;Suarez-Garcia et al., 2002).During wildfire emissions, anhydrites were found to be pronounced in the smoke microstructure, well in line with increased concentration of SO 4 -2 ions observed during a recent smoke event (Popovicheva et al., 2014a).
Measured SO 4 -2 , Ca 2 + , Cl -, and PO 4 3-ions (Table 3) may confirm the calcium salt constituents.Higher concentrations of SO 4 -2 ions in the flaming than in the smoldering phase correlate to increasing evaporation of SO 2 with temperature and following gas-to particle conversation.Group S-rich identified in Siberian biomass smoke in both combustion phases contained particles of exclusively S, C, and O with small inclusions of Cl and Na.Fig. 4.6 shows the typical SEM micrograph and EDX spectra of particles from debris flaming fires, demonstrating the soot morphology, with sulfur approaching 10 wt% on average and up to 35 wt% in the debris flaming and smoldering phases, respectively.Sulfates were likely produced in form of sulfuric acid on the soot surface.Formation of ammonium sulfates can also be proposed if the salt formation did not change the morphology towards to euhedral shape, since SO 4 -2 and NH 4 + ions were measured with the highest concentrations in the pine flaming phase.
Finally, the Group N-rich was identified only in debris smoke with low abundance of 2% (not shown in Table 1).Its average composition in the smoldering phase was C 67 O 9 N 24 , while in some particles the N content approached 38 wt%.The origin of nitrates in natural biomass and the mechanism of nitrate formation due to gas-to particle interaction of acidic NO and NO 2 gases during combustion is similar to those of sulphates (Suarez-Garcia et al., 2002).Observation of NO 2 -and NO 3 -ions at higher concentrations in the debris flaming phase, Table 3, confirm the nitrate and nitrite composition of Group N-rich.Likely they formed ammonium nitrates because no alkaline or alkaline-earth ions were found as exchangeable elements in particles of this group.
In numerous BB studies soot and organic particles were found with potassium inclusions (Liu et al., 2000;Osan et al., 2002;Posfai et al., 2003;Popovicheva et al., 2014a), while the K + ion is widely accepted as tracer of wood burning (Caseiro et al., 2009;Engling et al., 2011;Cheng et al., 2013).Its concentration in PM may account between 0.5 and 6% depending on biomass type (Reid et al., 2005).In Siberian pine and debris smoke, the contribution of K + to PM mass was found to be 0.8 and 0.2% in the flaming and smoldering phase, respectively.
Potassium was identified as a dominant element in almost all groups of wood combustion particles, thus serving as a marker element (Osán et al. 2002).For example, in savanna open fires elevated numbers of K, Cl-containing groups were found (Liu et al., 2000).However, we found low abundance of K in smoke particles (Fig. 2).In the flaming phase K was distributed in 6 and 4% of particles in pine and debris smoke, respectively, and did not produce a separate group.Analysis of K-containing individual particles showed that half of them contained sulfur in form of potassium sulfates (Fig. 4.7), in accordance with measured K + and SO 4 2-ions.The other half was mixed with Ca, Cl, and aluminosilicates (Fig. 4.5).

Time Evolution
Following the fire emission and smoke dispersion in the chamber, the morphology and composition of aerosols are changing due to various physical and chemical processes, including coagulation, interaction with gaseous species, and deposition.Smoke evolution was accompanied by continuous decrease in the total PM mass almost due to deposition and wall losses in a LAC, in relating with decreasing the TC fraction (Table 2).In the flaming phase a noticeable reduction of the Group Soot abundance during 48 h, from 87 to 61% for pine smoke, was observed alongside an increase of the fly ash groups fraction.This transformation correlated with an increase of the OC/EC ratio, for pine PM 10 particles by a factor up to 8 (Table 2).This pronounced effect likely occurred due to condensation of gaseous volatile organic species on the soot particles when smoke was cooling and dispersed across the large chamber without air perturbations such as wind or ventilation.In regional haze dominated by smoke, the organic species could increase particle mass by up to 40%, with roughly half due to condensation near the fire source and the remainder from other processes during long term aging (Reid et al., 1998).
Moreover, during aging the inorganic smoke constituents undergo a complex set of transformations.Aged regional smoke particles showed a significant enrichment of species associated with secondary aerosol production in the form of sulfates and ammonium (Reid et al., 1998).In the flaming phase the abundance of Group S-rich increased 7 times in aged (48 h evolution) smoke.A new Group, K,Cl-rich, with a fraction of 7% of total particle number appeared, well in accordance with an increased abundance of K, Cl, and S in smoke particles (Fig. 2).In this group 60% of all particles were found in form of KCl (Fig. 4.8), 20% consisted of K 2 SO 4 mixed with Ca and Cl, and another 20% contained only Cl.This finding confirms the assumption of Liu et al. (2000) that potassium organically bound in fluids of vegetation can evaporate during burning, and further oxidation leads to nucleation and condensation of potassium salt particles in smoke.Elevated concentrations of K + and SO 4 2-ions in aged smoke in comparison with fire-emitted particles confirm such way of potassium salts formation, well in line with observations during BB episodes (Engling et al., 2011;Tao et al., 2013) and wildfire events (Agarwal et al., 2010;Popovicheva et al., 2014a).
During aging of smoke produced in the smoldering phase, the concentration of quasi-liquid tar in smoke particles significantly decreased, while the abundance of Group Organic particles increased, for debris smoke from 69 to 82%.Group K,Cl-rich did not appear after 48 h evolution, while the abundance of Group S-rich remained nearly unchanged, emphasizing that prominent transformations occur preferably following high-temperature combustion.Since condensation of vaporized compounds takes place in the smoldering phase already during the earliest stages of particle formation, we observed nearly unchanged abundance of all elements and particle groups in the aged smoke.

SUMMARY AND CONCLUSIONS
Microscopic and chemical characterization of multicomponent aerosols emitted by small-scale fires of Siberian biomass allowed the quantification of smoke microstructure at the individual particle level.The application of cluster analyses was extended to biomass burning by combining microscopic measurements with bulk chemical characterization.The comprehensive analysis of PM 10 and PM 2.5 smoke revealed the general and specific properties of carbonaceous and fly ash particles emitted from smoldering, open flaming, and mixed fires during controlled repeated burns of Siberian pine wood and coniferous debris, thus supporting the exploration of biomass burning aerosol composition.
Carbon chemistry dominated the particle composition.C and O from vaporized organics comprised the chain soot agglomerates and roughly spherical organic particles in Group Soot and Organic from flaming and smoldering burns, respectively.Inorganic constituents evaporated from Siberian biomass produced internally/externally mixed fly ash in Group Ca-, Si-, and Fe-rich with significantly lower abundance.Experimental fires of regional biomass performed under natural conditions showed significantly higher abundance of Group Ca-, Si-, Fe-, Mn-, and K-rich due to prominent impact of dust minerals evolved by air convection from soil (Popovicheva et al., 2014a).Thus, the approach developed in this study helps to distinguish the impact of soil on biomass burning aerosol composition.
Sulfates and nitrates formed from gas-to-particle reactions of SO 2 , SO 3 , NO, NO 2 , and NH 3 gases produced Group Sand N-rich.Predominantly in the flaming phase, calcium and potassium combined with carbonates and sulfates comprised the inorganic salts.During time evolution of smoke volatile inorganic compounds condensed as potassium chlorides and sulfates into the newly formed Group K,Cl-rich which confirms the transformation of smoke microstructure during aging.This finding may provide an advanced tool for receptor models which favor potassium as smoke tracer for apportionment of biomass burning emissions (Reid et al., 2005).
The biomass type (pine wood vs debris) did not play a big role in terms of influencing the smoke microstructure.Ca, Cl, S, and Mg were more frequently distributed elements in pine smoke than debris smoke, with low abundance of the Group N-rich being specific for debris smoke.In prescribed burns of various Siberian forest types (mixed and coniferous) trace elements and carbonaceous constituents of smoke emissions were found to be almost the same under similar weather conditions (Samsonov et al., 2012).Thus, we conclude that the combustion phase is the key factor governing smoke microstructural features.
Cluster analysis of individual particles performed for specific sources, in combination with total emission inventories, may provide characteristic source profiles based on abundances of aerosol groups in different emission sources.This information is of great importance for assessing source strengths in source apportionment studies.Thus, the groups with K, Cl, and Na have been found as biofuel micromarkers to discriminate between diesel and biofuel exhaust emissions (Popovicheva et al., 2014b).Large intense fires of dry vegetation are characterized by high temperatures and oxygen limit, and produce soot particles with mostly chain-like structures (Reid et al., 1998).Therefore, the extending abundance of Group Soot in emission smoke may reveal a micromarker of open flaming during forest wildfires, well in accordance with low OC/EC ratio typical for high-temperature source of combustion.In opposite, the large emission of OC versus EC in the low-temperature phase produces almost organic particles, allowing the assignment of Group Organic to a micromarker of smoldering wildfires.Thus, the quantification of smoke microstructure by chemical micromarkers emphasizes the differences in aerosol properties from various Siberian wildfire emission types.Since this approach has also indicated the micromarker of large scale wildfires at the regional scale after long -range transport (Diapouli et al., 2014), it can help in assessment of dangerous sources of Arctic pollution by Siberian boreal forest wildfires, thus decreasing the uncertainties in estimates of biomass burning impacts on the atmosphere and climate.

Fig. 1 .
Fig. 1.Panorama of an impaction spot of a) pine flaming smoke particles and b) debris smoldering particles, c) typical soot agglomerates and d) quasi-liquid tar at the center of spot.

Fig. 2 .
Fig. 2. Abundance of trace elements in smoke particles from a) flaming and b) smoldering fires: Fresh pine, debris and aged pine (48 h of evolution) smoke.

Table 2 .
PM 10 and PM 2.5 mass with standard deviation, and carbon fractions in pine and debris smoke from flaming, smoldering, and mixed fires.Table.3.Average ion concentrations in PM 2.5 with standard deviation in pine and debris smoke of flaming, smoldering, and mixed fires smoke, in %.